1
MORPHOLOGICAL STRUCTURE AND PHONOLOGICAL DOMAINS IN SPANISH DENOMINAL DERIVATION
RICARDO BERMÚDEZ-OTERO University of Manchester
Forthcoming in: Colina, Sonia & Fernando Martínez-Gil (eds), Optimality-theoretic studies in Spanish phonology. Amsterdam: John Benjamins.
1 0. Introduction In Spanish, nominal words fall into a number of inflectional classes, each characterized by a particular stem formative:2 thus, o-stems bear the formative /-o/, a-stems bear the formative /-a/, and e-stems bear the formative /-{e,Ø}/ (less frequently, /-e/).3 Together, these three core classes comprise the vast majority of Spanish nouns and adjectives; athematic stems, which lack stem formatives altogether, are comparatively rare.
(1) SG PL gender gloss a. o-stems libr-o libr-o-s M ‘book’ niñ-o niñ-o-s M ‘boy’ tont-o tont-o-s M ‘silly’ re-o re-o-s M/F ‘convict’ man-o man-o-s F ‘hand’ b. a-stems libr-a libr-a-s F ‘pound’ niñ-a niñ-a-s F ‘girl’ tont-a tont-a-s F ‘silly’ ingles-a ingles-a-s F ‘English’ pianist-a pianist-a-s M/F ‘pianist’ problem-a problem-a-s M ‘problem’ c. e-stems libr-e libr-e-s M/F ‘free’ lápiz-Ø lápic-e-s M ‘pencil’ cruz-Ø cruc-e-s F ‘cross’ cruc-e cruc-e-s M ‘crossing’ inglés-Ø ingles-e-s M ‘English’ común-Ø comun-e-s M/F ‘common’ inmun-e inmun-e-s M/F ‘immune’ hindú-Ø hindú-e-s M/F ‘Hindu’ rey-Ø rey-e-s M ‘king’ d. athematic menú menú-s M ‘menus’ stems jersey jersey-s M ‘pullover’ esnob esnob-s M/F ‘snob’ clip clip-s M ‘paper clip’ virus virus M ‘virus’ brindis brindis M ‘toast’
Establishing the exact inventory and underlying distribution of nominal stem formatives is a key task for Spanish morphophonology, for it is impossible to analyse the phonotactics or the metrical system of the language without making crucial assumptions about the phonological behaviour of these elements. Discussions of Spanish stress assignment illustrating this link 2 include Harris (1983: 114-116, 1992: 75-76), Oltra-Massuet and Arregi (2005: §3.1), and Roca (1988: 416, 1991: note 11, 2005: 357-358), among many others. On the surface, nominal stem formatives seem never to occur before derivational suffixes. This is one of their most salient properties.
(2) base derivative man-o ‘hand’ man-az-a *man-o-az-a ‘hand.AUG’ problem-a ‘problem’ problem-ón-Ø *problem-a-ón-Ø ‘problem.AUG’ nub-e ‘cloud’ nub-os-o *nub-e-os-o ‘cloudy’
In principle, there are two possible approaches to the pattern exemplified in (2). One possibility is to posit some sort of morphotactic restriction preventing nominal stem formatives from occurring inside derivational suffixes, as in (3a). In this view, the stem formative of the base is absent from the underlying phonological representation of the derivative, and the morphological structure of the derivative is faithfully reflected in its surface representation. I shall hereafter refer to analyses of this type as ‘purely morphological’.
(3) Purely morphological analysis of (2) a. Morphotactic restriction4 * SF DER where SF = stem formative DER = derivational suffix b. Phonological mapping UR man o man a a SR [ ma .no ] [ ma.na . a ] ‘hand’ ‘hand.AUG’
Alternatively, one can postulate a morphologically sensitive phonological process of deletion applying to unstressed stem-final vowels before vowel-initial suffixes. In (4a) I provide a preliminary statement of this process; the precise conditions listed in the structural description are motivated in §2.1 below; tableau (32) offers an optimality-theoretic implementation. In this analysis, which I shall henceforth describe as ‘morphophonological’, stem-final vowel deletion masks the underlying morphological structure of stem-based derivatives.
(4) Morphophonological analysis of (2) a. Stem-final vowel deletion σw V → Ø / __ stem suffix V (noniterative) b. Phonological mapping UR man-o man-o a -a SR [ ma .no ] [ ma.na . a ] ‘hand’ ‘hand.AUG’
The alternative presented by these two approaches has of course not gone unnoticed in the literature (e.g. Pensado 1999: 4461). A similar question arises in Italian (Montermini 2003: §2), where Peperkamp (1995) adopts a purely morphological analysis along the lines of (3), whereas Scalise (1983: 74-78, 287ff.; 1994: 105, 151ff.) posits a rule of vowel deletion analogous to (4a). It is therefore rather surprising to observe that the choice between (3) and (4) has scarcely been discussed in mainstream generative work on Spanish. Within this tradition, most linguists have followed Harris (1983, 1985, 1991, 1992, 1996, 1999) in 3 assuming that Spanish denominal derivational suffixes attach to bases obligatorily lacking stem formatives: e.g. Oltra-Massuet and Arregi (2005: notes 8, 29, 30, 33, 38, 45), Roca (1990: 135, 1991: 604, 2005: 358).5 The problem calls for closer scrutiny since, as I pointed out above, one cannot describe Spanish stress or Spanish phonotactics without making crucial assumptions about the behaviour of stem formatives. In this chapter I examine the issue from the viewpoint of Stratal Optimality Theory (StrOT), a version of Optimality Theory (OT) in which morphology-phonology interactions are modelled by means of cyclicity and level segregation, as in Lexical Phonology and Morphology (LPM): see e.g. Bermúdez-Otero (1999, 2003, forthcoming a), Bermúdez-Otero and McMahon (2006), Booij (1996), Inkelas and Zoll (2005), Kiparsky (1998, 2000, 2003), and Orgun (1996), among others. It turns out that, in a highly constrained version of StrOT allowing no more than three phonological levels, the purely morphological analysis of Spanish nominal stem formatives shown in (3) leads to insurmountable difficulties: it creates a stratification paradox involving the two well-known phonological phenomena of diphthongization and depalatalization. In consequence, StrOT forces one to countenance the long-neglected morphophonological hypothesis set out in (4). Gratifyingly, this turns out to be the correct option: morphological arguments independent from diphthongization and depalatalization consistently favour (4) over (3). This result provides strong support for StrOT, which emerges from the trial as an empirically adequate, highly restrictive, and heuristically powerful model of grammar. The chapter is organized as follows. Section 1 introduces the key principles of StrOT and shows how their application to Spanish requires the postulation of a morphophonological process of stem-final vowel deletion. In section 2, I examine and refute Harris’s arguments for the purely morphological analysis of nominal stem formatives: section 2.1 shows that stem-final vowel deletion is motivated independently by the behaviour of verb stems; section 2.2 examines denominal derivation with consonant-initial suffixes; and section 2.3 adduces new evidence to disprove Harris’s claim that, in words such as virus ‘virus’ and brindis ‘toast’, the final /Vs/ string is an exotic ‘word marker’ (cf. (1d)). Significantly, the data presented in section 2.3 reveal a previously unrecognized contrast between pseudoplural nouns (e.g. sg. Sócrat-e-s ‘Socrates’, pl. Sócrat-e-s, dim.sg. Socrat-it-o) and nouns with athematic stems ending in /s/ (e.g. sg. virus, pl. virus, dim.sg. virus-it-o). In addition, the behaviour of pseudoplural bases casts new light on the relative rôles of suffixes and infixes in Spanish diminutive formation. Section 3 comments on the significance of our results.
1. The view from Stratal Optimality Theory 1.1. Principles StrOT aims to solve the problem of phonological misapplication in OT assuming no correspondence relationships other than input-output faithfulness; it therefore rejects both output-output correspondence (e.g. Benua 1997) and sympathy (McCarthy 1999). In addition, StrOT generates nonparadigmatic opacity by the same means as cyclic effects (Bermúdez- Otero 2003: §8; Bermúdez-Otero and McMahon 2006: §3.2). The theory relies on three basic concepts: those of domain, cycle, and level (Bermúdez-Otero forthcoming a: ch. 2). Phonology may be thought of as a function P mapping input representations onto the corresponding outputs. In OT, P is modelled as a pass through GEN and EVAL:
(5) P(x) = Eval(Gen(x))
In StrOT, however, P applies recursively: certain constituents in the morphosyntactic structure of a linguistic expression (or, alternatively, certain operations in its morphosyntactic derivation) define domains for P; each such domain provides the input for an application of P. Bermúdez- Otero (forthcoming a: ch. 2) argues that the relationship between grammatical structure and 4 phonological domains is in fact one of simplification: each phonological domain is exactly coextensive with some grammatical constituent, but not every grammatical constituent defines a phonological domain (cf. Inkelas 1990, Orgun 1996).6 Within the nested hierarchy of phonological domains associated with any linguistic expression, P applies cyclically in the sense of Chomsky and Halle (1968: 15). However, domains associated with different types of grammatical constituent (or, alternatively, with different types of grammatical operation) may invoke different rankings of CON: it is in this sense that domains are said to belong to different ‘levels’. Thus, for example, if a linguistic expression e has the domain structure shown in (6a), where the subscript indices denote the level to which each domain belongs, the surface representation of e will be specified by the composite function (6b).
(6) a. e = [c [b [a x]] [b [a y] z]]
b. P(e) = Evalc(Gen( Evalb(Gen( Evala(Gen(x)) )), Evalb(Gen( Evala(Gen(y)), z)) ))
In principle, this theoretical programme could be implemented in several ways. Bermúdez- Otero (forthcoming a) is primarily concerned with providing a highly restrictive version of StrOT that will curb the complexity of opacity effects and facilitate the acquisition of opaque interactions. This is best achieved in two ways: by limiting the number of phonological levels within the grammar, and by constraining the ascription of grammatical categories to phonological levels. Therefore, following a long tradition of research both within and outside LPM, Bermúdez-Otero (forthcoming a) adopts the hypothesis that, universally, grammars distinguish just three phonological levels: the stem level (SL), the word level (WL), and the phrase level (PL); see also Kiparsky (1998, 2000, 2003). Within this version of StrOT, the relationship between morphological constructions and phonological levels is regulated by principles that refer to three key morphological categories: root, stem, and word. In brief, a stem is defined as a form that can provide the base for an inflectional operation; roots cannot be inflected without first undergoing root-to-stem conversion (overtly or covertly), whereas words are fully inflected (i.e. syntactically free).
(7) Morphological categories inflectable? fully inflected? root stem word ()
On this basis, the grammar is set up in such a way as to enforce the following correspondences between grammatical constructions and phonological domains (for detailed discussion, see Bermúdez-Otero forthcoming a: ch. 2):
(8) a. Roots do not define phonological domains. b. A phonological domain associated with an operation of root-to-stem derivation must be stem-level. c. Every morphological word defines a word-level domain. d. The highest phrasal category in the linguistic expression defines a phrase-level domain.
5
At this point, it is important to note that terms such as ‘stem-level’ and ‘stem-based’ are not synonymous: the former refers to the phonological properties of an affix; the latter, to its morphological subcategorization requirements. Thus, an affix is stem-level if it defines phonological domains that invoke the stem-level constraint hierarchy; it is stem-based if it attaches to stems. Thus, an affix may be root-based and stem-level, stem-based and stem-level, stem-based and word-level, etc.; by (8b), however, it cannot be root-based and word-level. In terms of learnability, the austere implementation of StrOT outlined in this section enjoys significant advantages. First, the phonological domain structure of linguistic expressions remains relatively simple; there is no proliferation of cycles (cf. Orgun 1996).7 Secondly, learners with access to the grammatical structure of linguistic expressions can easily discover their phonological domain structure.
1.2. A stratification paradox? However, if scholars like Harris, Roca, and Oltra-Massuet and Arregi are right in their view of Spanish nominal stem formatives, then the StrOT model I have just outlined proves too austere: this section will show that, in a system with no more than three phonological levels, positing underlying representations along the lines of (3b) leads to a stratification paradox. The problem arises over the stratal ascription of two well-known phonological phenomena: so-called ‘diphthongization’ and ‘depalatalization’. For the sake of convenience I shall retain these traditional labels here, although, as we shall see presently, in neither case does a phonological process derive surface alternants synchronically from a common underlier.
1.2.1. Diphthongization So-called ‘diphthongization’ affects a sizable but idiosyncratic set of lexical items that display an alternation between the diphthongs /je, we/ and the mid vowels /e, o/: the diphthongs appear in tonic syllables, the mid vowels elsewhere. See Cole (1995: §6.2) for a survey of LPM analyses, and Eddington (2004: §6.1) for a summary of psycholinguistic approaches.
(9) a. Alternating items ’o] ‘blacksmith-שer-o] ‘iron’ [er-e ] ’o] ‘doorman-שt-e שt-a] ‘door’ [poשpwe] [djent-e] ‘tooth’ [dent-a l-Ø] ‘dental’ ’t-a l-Ø] ‘mortalשt-e] ‘death’ [moשmwe] ’je֙-o] ‘blind’ [ e֙-eᾩa -Ø ] ‘blindness ] [nwe ਫ-o] ‘new’ [noਫ-eᾩa -Ø] ‘novelty’
b. Nonalternating items ’o] ‘basket weaver-שest-a] ‘basket’ [ est-e ] ’o] ‘well digger-שpo -o] ‘well’ [po -e ] k k [a iᾩe nt-e] ‘accident’ [a iᾩent-a l-Ø] ‘accidental’ ’a l-Ø] ‘choral-שo] ‘chorus’ [ko-שko] ’k-eᾩa -Ø] ‘stubbornnessשk-o] ‘stubborn’ [teשte] [mo -o] ‘young’ [mo -eᾩa -Ø] ‘youth’
6
I assume that alternating items have two lexically listed allomorphs, whereas nonalternating items have a single underlying representation:
(10) a. Alternating items b. Nonalternating items o t a po oש p / we - / / - /
‘door’ ‘well’
In this view, diphthongization involves phonological selection between listed allomorphs (e.g. Kager 1996; Mascaró 1996; McCarthy 2002: 152-5, 183-4; Rubach and Booij 2001): the phonological constraint hierarchy preserves the quality of input vowels, but, when given the choice, favours diphthongs in tonic syllables and monophthongs elsewhere. This effect is illustrated in tableau (11), where I use TONIC→DIPHTHONGAL as an informal label for whatever context-sensitive markedness constraints favour diphthongs in tonic syllables, whilst *DIPHTHONG is a context-free markedness constraint requiring pure vowels.
(11)
phonology morphology →
IPHTHONG IPHTHONGAL ONIC DENT D I T input output *D * !* [ta.שpwe ] /t-aשpo/ !* [ta.שpo ] t-aש{p{o,we * [ta.שL [pwe /t-aשL /pwe * !* [ta.שpo ] * (*) !* [oש.te .שpwe] /o-שt-a-eשL /po (*) [oש.te .שL [po o-שt-a eש{p{o,we !* (*) [oש.te .שpwe] /o-שt-a-eשpwe/ (*) !* [oש.te .שpo] [pwe . o] *! * po -o /po -o/ L [po . o] * * (*) !* [oש.pwe. e ] /o-שo /po -o-e-שpo -o e (*) [oש.L [po. e
For our present purposes, the crucial point is that the selection of the diphthongal allomorph under primary stress overapplies in the presence of certain stress-attracting affixes, such as superlative -ísim-o and the so-called ‘evaluative’ suffixes, i.e. diminutive -(ec)it-o, augmentative -az-o, -ón-Ø, etc.
(12) a. Base [bwen-o] ‘good’ b. Normal nonapplication [bon-da -Ø] ‘goodness’ c. Overapplication [bwen-փ sim-o] ‘best’ [bwen-a -o] ‘good.AUG’ [bwen-o n-Ø] ‘good.AUG’ [bwen-(e )փ t-o] ‘good.DIM’8 7
In StrOT, the overapplication effect in (12c) shows that the domain of diphthongization excludes superlative -ísim-o as well as the evaluative suffixes. Accordingly, we must infer that diphthongization is a stem-level process; that suffixes such as -er-o, -al-Ø, and -(i)dad-Ø are stem-level; and that superlative -ísim-o and the evaluative suffixes are word-level.9
(13) portero puertita o o t a it aש o p שt a eש p domain structure WL SL we - - - WL SL we - - ta.שo pwe ש.te .שSL (Diphthongization) po tփ .ta.שo pweש.te .שWL po ‘doorman’ ‘door.DIM’
1.2.2. Depalatalization It has long been known that Spanish does not tolerate palatal consonants in domain-final position (Alonso 1945: 96-97). In this chapter the term ‘depalatalization’ will refer to this phonotactic restriction. The precise nature of the markedness constraint responsible for it need not concern us here; below I shall use *CODA/PAL for convenience.
(14) a. Palatal in onset position b. Nonpalatal in domain-final position ’(to disdain’ [des.ᾩe n] ‘disdain (N‘ [שdes.ᾩe.ْa ] [don. e .ѭa] ‘lass’ [don. e l] ‘lad’
The literature on the subject is abundant: for Harris’s last statement on the subject, see Harris (1999); for a recent optimality-theoretic treatment, see Lloret and Mascaró (2005). It should be noted that, synchronically, the alternants shown in (14) are not derived from a common underlier (Hualde 1989: §6; cf. Lloret and Mascaró 2005). Psycholinguistic experimentation indicates that these alternations are unproductive: native speakers fail to extend them to neologisms (Pensado Ruíz 1997; Eddington 2004: §3.4). This conclusion agrees well with the internal evidence. Observe that, as far as their morphology is concerned, desdén-Ø and doncel-Ø are perfectly ordinary e-stem nouns, as shown by their plural forms: desden-e-s, doncel-e-s; see (1c). After illegal domain-final sequences, however, singular e- stem nouns select the /-e/ allomorph of the stem formative; the /-Ø/ allomorph appears only after permissible domain-final strings (Harris 1999; Bermúdez-Otero forthcoming a: ch. 4).10 By implication, no singular e-stem noun or adjective ends in a sequence that is forbidden domain-finally in the core vocabulary: if the /-e/ formative is absent in that environment, then it is also absent in the plural, showing the stem to be athematic rather than a member of the e- class. Examples include partially assimilated loans such as sg. cli[p] ‘paper clip’ ~ pl. clip-s, not *clip-e-s.11 There is thus an altogether natural and predictable relationship of implication between phonological markedness and morphological structure in Spanish nominal morphology: speakers know tacitly that, if a singular noun or adjective is phonotactically deviant, then it cannot belong to one of the core native stem classes and must consequently be athematic.12 However, this entails that a hypothetical e-stem noun derived from the root ,desdeْ-/ would surface as *desde[ْ]-e, not desdé[n]-Ø: see tableau (15).13 Synchronically/ therefore, the alternation between the verb desdeñ-a-r and the noun desdén-Ø is suppletive. 8
(15)
AL
phonology /P -Place -C
morphology -V -V ODA AX EP INAL DENT D M F *C input output I [des.de .ne] *! * * !* [desden-e/ [des.de ْ/ actual UR [des.de n] *! des.de .ne desden-{e,Ø} [ ] *! * * !* [L /desden-Ø/ [des.de ْ L [des.de n] * [des.de .ْe] * !* [desdeْ-e/ [des.de ْ/ counterfactual UR [des.de n] *! * des.de .ْe * !* [ ] {desdeْ-{e,Ø !* [desdeْ-Ø/ [des.de ْ/ [des.de n] *!
Of course, even though the alternations in (14) are lexically listed rather than synchronically derived, depalatalization remains perfectly robust as a principle of allomorph selection and as a static phonotactic restriction, as shown by the evidence of loan adaptation. In the present context, the interest of depalatalization lies in the fact that it applies not only in phrase-level and word-level domains, but also in sublexical domains which can be proved to be stem-level. The crucial piece of evidence is this: word-final consonants never have palatal alternants before word-level suffixes such as augmentative -az-o or diminutive -(ec)it-o.
(16) a. Base b. Word-level derivative [don. e l] ‘lad’ [don. e.lփ .to] *[don. e.ѭփ .to] [pje l] ‘leather’ [pje.l(e. )փ .ta] *[pje.ѭ(e. )փ .ta]14
However, if palatals were permitted domain-finally at the stem level, then this type of alternation would be possible. Consider, for example, the derivation of the singular and the diminutive singular forms of a hypothetical e-stem noun /kantiѭ-{e,Ø}/.15 With depalatalization active only at the word level, the result would be an impossible alternation between [kan.tփ l] and [kan.ti.ѭփ .to].
(17) a. Incorrect derivation SG DIM.SG domain structure WL SL kantiѭ-{e,Ø} WL SL kantiѭ-{e,Ø } it-o SL (*CODA/PAL ranked low) kan.tփ ѭ kan.tփ ѭ WL (*CODA/PAL ranked high) kan.tփ l kan.ti.ѭփ .to b. Correct derivation domain structure WL SL kantiѭ-{e,Ø} WL SL kantiѭ-{e,Ø} it-o SL (*CODA/PAL ranked high) kan.tփ .ѭe kan.tփ .ѭe WL (*CODA/PAL ranked high) kan.tփ .ѭe kan.ti.ѭփ .to
We must conclude that palatals are banned domain-finally already at the stem level. 9
Note that our analysis correctly predicts that depalatalization cannot cause alternations between the singular and plural forms of e-stem nouns. As noted by Harris (1999: 69) and Bermúdez-Otero (forthcoming a: ch. 4, Spanish has a morphological rule, stated informally in (18) below, which requires e-stem nominals to take the nonnull allomorph of their stem formative before the plural suffix. The presence of -e- in plurals such as [indu- e-s] ‘Hindus’ [and [re ٍ-e-s] ‘kings’ fulfils a morphological requirement, not a phonotactic one: *[indu- Ø-s and *[rej-Ø-s] are phonotactically impeccable; cf. [aw.to.ਫu s] ‘bus’, [xe.su s] ‘Jesus’, [sejs] ‘six’, [bejs] ‘beige’. In the same way, constrasts such as [indu- e-s] ‘Hindus’ vs [menu- s] menus’ or [re ٍ-e-s] ‘kings’ vs [xerse j-s] ‘pullovers’ are symptomatic of a morphological‘ distinction (that between e-stems and athematic stems) rather than the operation of a phonological process. See further Bonet (this volume) and Colina (2003a, 2003b).
(18) NSF PL -{e,Ø} → -e / ___ -s
Accordingly, plural e-stem nouns and adjectives take the /-e/ formative regardless of the shape of the root. However, if the root ends in a palatal, then the singular form will also select the /-e/ allomorph in order to satisfy *CODA/PAL: see tableau (15).
(19) a. Palatal-final root: ‘street’ SG PL domain structure WL SL kaѭ-{e,Ø } WL SL kaѭ-e s SL ka .ѭe ka .ѭe WL ka .ѭe ka .ѭes b. Non-palatal-final root: ‘lime’
domain structure WL SL kal-{e,Ø} WL SL kal-e s SL kal ka .le WL kal ka .les
In this sense, depalatalization does not overapply in plural e-stems. Although one can use output-output correspondence to account for the absence of alternations such as *[kan.tփ l ~ kan.tփ .ѭes], such a solution relies on a misrepresentation of the morphological facts (pace Lloret and Mascaró 2005).16
1.2.3. The paradox and its solution So far, this section has established the following points: 1. diphthongization applies at the stem level; 2. palatal consonants in domain-final position are prohibited at all levels; 3. superlative -ísim-o and the evaluative suffixes are word-level. In this light, consider now the following data:
(20) a. Base b. Stem-level derivative c. Word-level derivative [Ø ] [kweѭ-a -o-שkwe ѭ-o] [koѭ-a ] ‘neck’ ‘collar, necklace’ ‘neck.AUG’
Paradigms of this type are interesting because they show three phenomena occurring simultaneously: 10
1. the root ends in a palatal consonant; 2. the root is subject to the diphthongization alternation; 3. as per (2) above, the stem formative of the base fails to surface in derivationally related forms. For our purposes, the crucial datum is augmentative cuell-az-o: (20c). It shows overapplication of diphthongization, which is entirely expected since the suffix -az-o is word-level. The question is: how should we account for the fact that the root-final consonant fails to depalatalize? Let us first assume that the purely morphological approach outlined in (3) gives us the correct analysis of the behaviour of Spanish nominal stem formatives. If so, the morphological structure of cuell-az-o will be the following:
(21) k{o,we}ѭ a o
This underlying representation creates a dilemma. If -az-o is indeed a word-level suffix, then the underlying /ѭ/ will be domain-final in the stem-level cycle and incorrectly undergo depalatalization: see (22a). If, in contrast, -az-o is already visible to the phonology in the stem-level cycle, then the palatal will be syllabified in onset position and escape depalatalization, but the root-vowel will incorrectly fail to diphthongize: see (22b). In other words, we have a stratification paradox: the augmentative construction can be neither stem- level nor word-level, yet it must be one of the two.
(22) a. -az-o is word-level b. -az-o is stem-level
domain structure WL SL k{o,we}ѭ a -o WL SL k{o,we}ѭ-a -o SL kwe l ko.ѭa . o WL *kwe.la . o *ko.ѭa . o17
And, recall, we cannot solve the problem by turning depalatalization off at the stem level because that would give rise to unattested alternations such as *[kan.tփ l ~ kan.ti.ѭփ. to]; see (17). The alternative hypothesis is that, in the output of the morphology, cuell-az-o has the structure shown in (23); the disappearance of the stem formative of the base is caused by the phonological process of stem-final vowel deletion described in (4a).
(23) k{o,we}ѭ-o a -o
If we take this view, the stratification paradox evaporates. In the stem-level cycle, the stem formative of the base provides an onset position for the root-final consonant, which accordingly evades depalatalization. In the word-level cycle, the palatal remains in the onset despite stem-final vowel deletion, but stress migrates to the augmentative suffix, causing diphthongization to overapply.
(24) domain structure WL SL k{o,we}ѭ-o a -o SL kwe .ѭo WL kwe.ѭa . o
We must conclude that nominal stem formatives do occur inside derivational suffixation. Thus, the internal logic of StrOT drives one inexorably to reject the view of Spanish nominal 11 morphology assumed by Harris, Roca, and Oltra-Massuet and Arregi. Insofar as the endings /-o/, /-a/, and /-e/ can precede derivational suffixes, they cannot be ‘word markers’ in the sense of Harris (1985), but must be genuine stem formatives: i.e. meaningless morphs added to the root in order to satisfy a ‘morphomic’ constraint on stem well-formedness (Aronoff 1994). Logically, if a denominal derivative is stem-based (as opposed to root-based), the derivational affix will attach outside the stem formative of the base. More generally, nouns and adjectives belonging to the three core classes (i.e. o-stems, a-stems, and e-stems) turn out to have exactly the same tripartite morphological structure as verbs:
(25) Morphological structure of major-category words (modulo athematic nominals)
a. Word Stem Root ___ SF INFL where SF = stem formative INFL = inflectional affixes b. verbs amábamos
Word Stem Root am -a -ba -mos love -SF -PRET.IPFV.IND -1PL ‘we loved’ c. nouns niños
Word Stem Root nփ ْ -o -s boy -SF -PL ‘boys’
In sum, the austere implementation of StrOT outlined in §1.1 makes very precise predictions about denominal word-formation in Spanish: in particular, it predicts that stem-based denominal derivatives are subject to the phonological process of stem-final vowel deletion formulated in (4a). In the remainder of this chapter I demonstrate that this prediction is, in fact, correct.
2. The morphological evidence Over the years, Harris has grounded his rejection of stem-final vowel deletion on three claims: 1. stem-final vowel deletion has no independent motivation (Harris 1991: footnote 9); 2. stem-final vowel deletion cannot account for the absence of nominal stem formatives before derivational suffixes beginning with a consonant (Harris 1983: 92, 147; 1996: 104); 3. stem-final vowel deletion cannot account for the behaviour of exotic ‘word markers’ such as -us in vir-us, or -is in brind-is (e.g. Harris 1992: 66, 75). In this section I show that each of these three claims is false (see also Bermúdez-Otero forthcoming a: ch. 4).
2.1. Stem-final vowel deletion in verb stems [o-փ n-aשo-e] ‘hero’ and its derivatives [eשDiscussing the contrast between the noun [e o-փ sm-o] ‘heroism’, Harris (1991: footnote 9) asserts that the final vowel ofשheroine’, [e‘ the base “is not an integral part of the stem” because “there is no independently motivated rule that would delete such a stem-final vowel”. This claim can be easily disproved, for the process of stem-final vowel deletion required to describe denominal derivation operates in exactly the same way in deverbal derivation and verbal inflection. As is well-known, Spanish verbs fall into three inflectional classes (‘conjugations’), characterized by the stem formatives /-a/, /-e/, and /-i/; these are subject to allomorphic patterns of gradation whose details need not concern us here (see e.g. Harris 1997: 547). The data in (26) show that verb stem formatives surface before consonant-initial derivational 12 suffixes. Thus, Harris must stipulate that verb stems are exempt from the putative morphotactic restriction in (3a).18
(26) a. infinitive b. nomen agentis c. participle caz-a-r ‘hunt’ caz-a-dor-Ø caz-a-d-o habl-a-r ‘talk’ habl-a-dor-Ø habl-a-d-o com-e-r ‘eat’ com-e-dor-Ø com-i-d-o beb-e-r ‘drink’ beb-e-dor-Ø beb-i-d-o hac-e-r ‘do’ hac-e-dor-Ø hech-o pon-e-r ‘put’ pon-e-dor-Ø puest-o abr-i-r ‘open’ abr-i-dor-Ø abiert-o dec-i-r ‘say’ dec-i-dor-Ø dich-o
Observe that the derived nomina agentis in (26b) cannot be analysed as departicipial formations in -or-Ø because they do not preserve the gradation and suppletion patterns of the corresponding participles. Needless to say, they are not deinfinitival either: rather, they are based on the verb stem. For example, caz-a-dor-Ø ‘hunter’ (pl. caz-a-dor-e-s) has the underlying representation shown in (27).
{e,Ø}-שN V ka -a do (27)
Consider now the augmentative suffix -ón-Ø. This can be added not only to nominal stems, but also to verb stems, from which it derives nomina agentis with jocular or derogatory connotations. By principle (8b), these constructions must be stem-based rather than root- based, since -ón-Ø is a word-level suffix: see §1.2.1, particularly (12). Moreover, the examples in (28b) are unambiguously deverbal, since they differ morphologically and/or semantically from the corresponding nomina actionis (Lázaro Mora 1999: 4673). Crucially, however, the augmentative nomina agentis lose the stem formative of the base before the initial vowel of the suffix.
(28) a. infinitive b. nomen agentis (AUG) c. nomen actionis acus-a-r ‘accuse’ acus-ón-Ø acuse, acusación fisg-a-r ‘pry’ fisg-ón-Ø fisgoneo trag-a-r ‘swallow’ trag-ón-Ø trago respond-e-r ‘answer’ respond-ón-Ø respuesta
The contrast between (26b) and (28b) indicates that, in deverbal derivation, the stem formative of the base surfaces before consonant-initial suffixes but is deleted before vowel- initial suffixes, exactly as predicted by (4a). In fact, stem-final vowel deletion applies in precisely the same way regardless of the syntactic category of the base. First, it does not iterate:
(29) a. Denominal derivation base derivative héro-e ‘hero’ hero-ín-a, *her-in-a ‘heroine’ bacala-o ‘cod’ bacala-ít-o, *bacal-it-o19 ‘cod.DIM’
b. Deverbal derivation base derivative pele-a-(r) ‘fight’ pele-ón-Ø, *pel-ón-Ø ‘quarrelsome’ mare-a-(r) ‘make dizzy’ mare-ón-Ø, *mar-ón-Ø ‘dizzying’
13
Secondly, stem-final vowel deletion does not aply to underlyingly accented vowels. In (30) I provide two examples from denominal derivation. Note that the bases café and papá are athematic stems, not e-stems, since their respective plurals are café-s and papá-s.20 In both, the final vowel must be underlyingly accented because final stress is a marked pattern for CVCV nouns (see e.g. Roca 1988: 398).
(30) a. base b. derivative UR SR
café ‘coffee’ kafe in-a [ka.fe.փ .na], *[ka.fփ .na] ‘caffeine’ 21 papá ‘Dad’ papa it-o [pa.pa.փ .to] ‘Dad.DIM’
The same phenomenon can be observed in verbal inflection. In the preterite imperfective indicative, for example, stress falls consistently on the stem formative. This pattern gives rise to prosodic configurations that are otherwise unattested in the native vocabulary. Notably, second-person plural forms have penultimate stress despite containing a falling diphthong in their final syllable: see (31b). This metrical pattern is impossible in nonverbal forms: e.g. [kom.bo j] ‘convoy’, not *[kom.bo j] (Harris 1983: §4.4.2, 1995: 870; Roca 1988: 398). This contrast indicates that the stress pattern of the preterite imperfective indicative is controlled by the morphology rather than the phonology: more specifically, an allomorphy rule assigns an underlying foot-head to the stem formative.22 As expected, the underlyingly accented stem formative fails to delete before the vowel of the tense, aspect, and mood marker.
(31) partíais t -փ -a -jsשa. UR Word Stem Root pa part -SF -PRET.IPFV.IND -2PL [tփ .ajs.שb. SR [pa ‘part.2PL.PRET.IPFV.IND’
In sum, we cannot concur with Harris (1991: footnote 9) when he asserts that there is no independent motivation for stem-final vowel deletion in Spanish denominal derivation: the same process, subject to identical restrictions, is at work in deverbal derivation and verb inflection. The following tableau provides a possible optimality-theoretic analysis; see Bermúdez-Otero (forthcoming a: ch. 4) for technical discussion.
(32) a. MAX-V&AdjSegMAX-V Assign one violation mark if two or more adjacent input vowels lack an output correspondent.23
b. MAX-V´ Assign one violation mark for every accented input vowel that lacks an output correspondent. c. ALIGN(suffix,onset) If an input vowel V is initial in a suffix attached to a stem, then assign one violation mark for every segment intervening between the output correspondent of V and the nearest preceding onset segment.24 14
(32) (cont.) d.
-V AX M AdjSeg
(suffix, onset) ´ -V& -V -V AX AX AX LIGN M M M A ma.no.a . a *! man-o Stem a -a ma.na . a 7 * ba.ka.la.o.փ .to **! bakala-o Stem it-o ba.ka.la.փ .to 7 * * ba.ka.lփ .to *! ** pa.pa.փ .to 7 * papa Stem it-o pa.pփ .to *! *
2.2. Denominal derivatives with consonant-initial suffixes Our analysis predicts that nominal stem formatives will be able to surface before derivational suffixes, provided that those suffixes are consonant-initial. Is this true? Harris (1983: 92, 147; 1996: 104) claims that it is not, adducing evidence from nomina qualitatis in -dad-Ø:
(33) a. base b. nomen qualitatis bell-o ‘beautiful’ bel-dad-Ø *bell-o-dad-Ø buen-o ‘good’ bon-dad-Ø *bon-o-dad-Ø herman-o ‘brother’ herman-dad-Ø *herman-o-dad-Ø
However, these constructions are altogether irrelevant to the matter at hand, since they are root-based rather than stem-based. Thus, the underlying representation of bon-dad-Ø is neither (34a) nor (34b), but (34c).
(34) a. Stem Stem b{we,o}n-o dad-{e,Ø}
b. Stem Stem b{we,o}n dad-{e,Ø}
c. Stem Root b{we,o}n dad-{e,Ø }
This is confirmed by three facts: 1. The relevant suffix has four allomorphs: -tad-Ø, -dad-Ø, -edad-Ø, and -idad-Ø. Of these, only -idad-Ø remains productive (Santiago Lacuesta and Bustos Gisbert 1999: 4536). In contrast, -dad-Ø is never found in neologisms, but only in words inherited from Latin: e.g. uēr-itāt-e-m > ver-dad-Ø ‘truth’, which first underwent intervocalic -t- lenition, and then syncope and apocope (Pharies 2002: 163). 2. Many nomina qualitatis in -dad-Ø have bound bases that do not exist as independent stems: e.g. frial-dad-Ø ‘coldness’, mortan-dad-Ø ‘mortality’, ver-dad-Ø ‘truth’; note the absence of *ver-o, *ver-a, *ver-e, or *ver-Ø ‘true’ in modern Spanish. These constructions can only be root-based. 3. Since the allomorph -dad-Ø attaches to roots, StrOT predicts that it will be stem-level, since word-level constructions cannot be root-based: see (8b). This prediction proves 15